Top AI Product

Stripe just dropped [Part 2 of their Minions blog series](https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents-part-2), and honestly, the numbers are kind of absurd. Over 1,300 pull requests merged every single week, entirely written by AI agents, with humans only stepping in for code review. Not “AI-assisted.” Not “co-piloted.” Fully autonomous, start-to-finish coding.

So what exactly are Minions? They’re Stripe’s internal, fully unattended coding agents. You kick one off through a Slack message or a ticket, and it comes back with a complete pull request that’s already passed CI. No hand-holding, no back-and-forth. The whole thing is built on a fork of [Block’s open-source Goose agent](https://github.com/block/goose), heavily modified to run without any human supervision. Each Minion spins up on an isolated AWS EC2 devbox that boots in about 10 seconds from a warm pool — pre-loaded with Stripe’s source code, services, and everything the agent needs to get to work.

What makes this especially interesting is how Stripe handles context. They’ve built an internal centralized MCP server called “Toolshed” that exposes nearly 500 tools to the agents. That means Minions can interact with internal systems, third-party SaaS platforms, linting, testing — you name it. They also introduced something called “blueprints,” which are hybrid orchestration flows mixing deterministic steps with open-ended agent loops. It’s a clever way to guarantee certain subtasks get done correctly while still letting the agent figure things out on its own for more complex problems.

The [Part 1 blog post](https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents) already generated a ton of discussion when it hit [Hacker News](https://news.ycombinator.com/item?id=46952149), and [Part 2](https://news.ycombinator.com/item?id=43086557) landed right back on the front page on February 19th. Twitter has been buzzing about it too, and for good reason — this is probably the most detailed public look at how a major tech company is actually deploying AI coding agents at scale. We’re talking about a company that processes over a trillion dollars in annual payment volume, running hundreds of millions of lines of mostly Ruby code. This isn’t a toy demo.

The thing that strikes me most is how practical the approach is. Minions run local linting before pushing, get one CI feedback cycle with autofixes, and then optionally self-remediate before a human even looks at the PR. It’s methodical, not magical. And the fact that they open-sourced the base agent through Goose means other companies could, in theory, build something similar — though Stripe’s secret sauce is clearly in the tooling and infrastructure around it.

If you’re interested in where enterprise AI coding is actually headed (not where people speculate it’s headed), the Stripe Minions blog is required reading.